Background Atopic dermatitis (AD) is a heterogeneous, chronic inflammatory skin disease linked to skin microbiome dysbiosis with reduced bacterial diversity and elevated relative abundance of Staphylococcus aureus (S. aureus). Objectives We aimed to characterize the yet incompletely understood association between the skin microbiome and patients' demographic and clinical cofactors in relation to AD severity. Methods The skin microbiome in 48 adult moderate‐to‐severe AD patients was investigated using next‐generation deep sequencing (16S rRNA gene, V1–V3 region) followed by denoising (DADA2) to obtain amplicon sequence variant (ASV) composition. Results In lesional skin, AD severity was associated with S. aureus relative abundance (rS = 0.53, p < 0.001) and slightly better with the microbiome diversity measure Evenness (rS = −0.58, p < 0.001), but not with Richness. Multiple regression confirmed the association of AD severity with microbiome diversity, including Shannon (in lesional skin, p < 0.001), Evenness (in non‐lesional skin, p = 0.015) or S. aureus relative abundance (p < 0.012), and with patient's IgE levels (p < 0.001), race (p < 0.032), age (p < 0.034) and sex (p = 0.012). The lesional model explained 62% of the variation in AD severity, and the non‐lesional model 50% of the variation. Conclusions Our results specify the frequently reported “reduced diversity” of the AD‐related skin microbiome to reduced Evenness, which was in turn mainly driven by S. aureus relative abundance, rather than to a reduced microbiome Richness. Finding associations between AD severity, the skin microbiome and patient's cofactors is a key aspect in developing new personalized AD treatments, particularly those targeting the AD microbiome.

Introduction: Microbiome amplicon sequencing data are distorted by multiple protocol-dependent biases, originating from bacterial DNA extraction, contamination, sequence errors, and chimeras. In particular, extraction bias is a major confounder in sequencing-based microbiome analyses, with no correction method available to date. Here, we suggest using mock community controls to bioinformatically correct extraction bias based on morphological properties. Methods: We compared dilution series of 3 mock communities with an even or staggered composition. DNA was extracted with 8 different extraction protocols (2 buffers, 2 extraction kits, 2 lysis conditions). Extracted DNA was sequenced (V1-V3 16S rRNA gene) together with corresponding DNA mocks. Sequences were denoised using DADA2, and annotated by matching against mock reference genomes. Results: Microbiome composition was significantly different between extraction kits and lysis conditions, but not between buffers. Independent of the extraction protocol, chimera formation increased with high input cell number. Contaminants originated mostly from buffers, and considerable cross-contamination was observed in low-input samples. Comparison of microbiome composition of the cell mocks to corresponding DNA mocks revealed taxon-specific protocol-dependent extraction bias. Strikingly, this extraction bias per species was predictable by bacterial cell morphology. Morphology-based bioinformatic correction of extraction bias significantly improved sample compositions when applied to different samples, even with different taxa. Conclusions: Our results indicate that higher DNA density increases chimera formation during PCR amplification. Furthermore, we show that bioinformatic correction of extraction bias is feasible based on bacterial cell morphology.